Kernel Parameters for Legato Networker (large environment). - HP UX

This is a discussion on Kernel Parameters for Legato Networker (large environment). - HP UX ; Hi Everyone, I'm looking for kernel parameter recipes for Legato Network in a large environment running on HP-UX 11.23. So far HP and Legato have been unable to provide and we've been iteratively tweaking until now which is a little ...

+ Reply to Thread
Results 1 to 5 of 5

Thread: Kernel Parameters for Legato Networker (large environment).

  1. Kernel Parameters for Legato Networker (large environment).

    Hi Everyone,

    I'm looking for kernel parameter recipes for Legato Network in a large
    environment running on HP-UX 11.23.

    So far HP and Legato have been unable to provide and we've been
    iteratively tweaking until now which is a little irritating to say the
    least. I'm not talking abot performance tuning... but guaranteeing
    stability and functionality under load.

    Dave


  2. Re: Kernel Parameters for Legato Networker (large environment).

    galooni wrote:
    > Hi Everyone,
    >
    > I'm looking for kernel parameter recipes for Legato Network in a large
    > environment running on HP-UX 11.23.
    >
    > So far HP and Legato have been unable to provide and we've been
    > iteratively tweaking until now which is a little irritating to say the
    > least. I'm not talking abot performance tuning... but guaranteeing
    > stability and functionality under load.


    *Which* kernel parameters are you "iteratively tweaking", why, how,
    etc.? I.e. kernel parameter "tweaking" is not a purpose in itself. You
    do it for a reason. "stability and functionality under load" are valid
    reasons, but without details, they are meaningless terms.

  3. Re: Kernel Parameters for Legato Networker (large environment).

    First of all the "tweaking" was historically with 11.11. We have
    satisfactory stability on 11.11.

    We want to avoid that scenario of iterative configuration now that we
    are going with 11.23, by implementing a kernel which is properly
    configured from the start.

    So what we're looking for is a /stand/system file which covers
    "stability and functionality". A recipe if you like.

    Here's the tunables we have modified so far on our new 11.23 systems...
    based upon our experiences with *11.11*...

    =======================
    tunable shmmax 17179869184
    tunable semmnu 260
    tunable ninode 16384
    tunable ncsize 21504
    tunable msgseg 32767
    tunable msgmnb 65535
    tunable msgmap 2048
    tunable msgtql 2046
    tunable maxvgs 64
    tunable maxdsiz 2147483648
    tunable max_thread_proc 512
    tunable executable_stack 1
    tunable dbc_min_pct 50
    tunable dbc_max_pct 90
    tunable nstrpty 60
    =====================


  4. Re: Kernel Parameters for Legato Networker (large environment).

    On 15 Sep 2005 05:00:53 -0700, "galooni" wrote:

    >First of all the "tweaking" was historically with 11.11. We have
    >satisfactory stability on 11.11.
    >
    >We want to avoid that scenario of iterative configuration now that we
    >are going with 11.23, by implementing a kernel which is properly
    >configured from the start.
    >
    >So what we're looking for is a /stand/system file which covers
    >"stability and functionality". A recipe if you like.
    >
    >Here's the tunables we have modified so far on our new 11.23 systems...
    >based upon our experiences with *11.11*...
    >
    >=======================
    >tunable shmmax 17179869184
    >tunable semmnu 260
    >tunable ninode 16384
    >tunable ncsize 21504
    >tunable msgseg 32767
    >tunable msgmnb 65535
    >tunable msgmap 2048
    >tunable msgtql 2046
    >tunable maxvgs 64
    >tunable maxdsiz 2147483648
    >tunable max_thread_proc 512
    >tunable executable_stack 1
    >tunable dbc_min_pct 50
    >tunable dbc_max_pct 90
    >tunable nstrpty 60
    >=====================



    I agree with Frank's last post. Much time can be wasted, or worse, by
    iteratively "tweaking". Not only that, but you can go blind :-)

    Where to begin? ...

    Why bump up the HFS in-core inode cache? Ditto for ncsize - do you
    have significant HFS operations?

    Your shmmax allows for individually large shared memory segments on a
    64-bit system (does Legato require this?), and implies that the system
    has the memory to back it up. So why is dbc_max_pct (expressed as a
    percentage of RAM) set to allow so large a filesystem buffer cache?

    There are no real tuning questions around maxvgs as it's simply a
    limit on the number of LVM volume groups. 64 is quite a few volume
    groups. Are disks separated like this for a reason eg PRM IO control?

    You'd have to do some additional work to use that configured 32-bit
    maxdsiz limit as it's larger than the default executable type data
    segment limit.

    With [the per-process limit] max_thread_proc bumped up from default, I
    think that's higher than the default for [the system-wide limit]
    nkthread. If the per-process limit was needed that high then you'd
    likely be seeing errors relating to the system-wide limit.

    Or put another way, the system is rather stable at default kernel
    param settings. It's only when you start introducing those pesky
    applications that things get dicy :-) Applications should provide
    some level of guidence for tunable settings based on their projected
    resource requirements ie as Frank suggested - reason based changes.

    In the absence of info from the app vendor, recording
    performance/resource metrics for the apps while under load would be a
    good way to go.


  5. Re: Kernel Parameters for Legato Networker (large environment).

    Thanks Eric for having replied.

    Okay, well we made plenty of reason based changes for which I coined
    the unfortunate term "iterative tweaking". Judging by the fact that the
    responses I've had (on the lists I've posted to) are mainly... "use the
    defaults and adjust if you *need* to....", I think we'll just stick to
    what we used on 11.11.

    It's a shame the vendors for "pesky" applications aren't a little more
    supportive. There are guidelines available for Legato Networker but
    their pretty sketchy.

    Anyway.... if for some reason we have to stray from our current
    recipe... I'll come back and post our changes with the appropriate
    ratilonale attached.

    I'm replying from home, so I'll answer your questions about the buffer
    cache, inode cache, vg limits sometime next week.... we're backing up
    a heck of a lot of data.... quite a few terabytes a day, and there's
    more than one backup server.


+ Reply to Thread